Introduction
Sentiment Analysis is the field of study in Computer Science which analyses and researches the opinions and sentiments of people they express on the internet. It can be an opinion on a movie, product, issue, event, company, etc. The main purpose of this subject is to study their opinions and give a general overview based on it.
However, there is a slight difference between Sentiment Analysis and Opinion Mining. Former analyses the sentiments of the people about an issue of their interest which directly effects them whereas the latter deals with the opinions of people about a product, service, or company.
Sentiment Analysis was first mentioned in the Dave, Lawrence and Pennock,2003.
This field saw a boom after 2000 when internet became a household name. People started to express their opinions on various social media platforms which made it necessary to analyze it in order to formulate what people think about a topic. Also, this field attracted various researchers since there were many challenging problems which needed to be solved. It poses many problems such as understanding natural language used by people to express their sentiments. It was easier to analyze it manually but became a problem when an automation was required to mine the large amount of data flowing on the internet.
Applications
Sentiment Analysis has wide area of application since almost every field requires to know the general trend among common people to make their future plans. Companies and Organizations wants to know public’s opinions about their products or services. People generally asked their friends and family when they wanted to get an opinion about a product. But, today people go online and check for product reviews on related blogs or websites. People wants to know the opinions of others before making an important decision. This dependence on each other makes this field important.
The growth of social media has boomed the internet with opinions. It becomes really difficult to chalk out an overview from such large amount of data. For instance, Elections see many people expressing their opinion about the candidates on Twitter and other social media platforms. Analysis of this makes it possible for the candidates to decide their next step. It helps the analytics to predict the outcome of the result based on general opinion of the public.
In the same way, Organizations and Companies analyzes the public’s opinions about their newly launched products or services. It helps them improve their services or add new features to their products based on what public requires. It was difficult for companies to know about public’s requirement so they conducted a public poll through either door-to-door survey or newspapers which took a lot of time and resources. The era of social networks has solved that problem for the companies but has posed a bigger problem of analyzing those opinions and sorting them out. Sentiment Analysis is the field which comes to their rescue.
Sentiment Analysis has spread to almost every field in recent years. Multi-national companies and even the Governments have now started analyzing trends and opinions from the social network for their major decisions. Some examples for the same are-
In (Liu et al, 2007), a sentiment model was proposed to predict the sales performance. In (Asur and Huberman,2010; Joshi et al., 2010;Sodikov, Parameswaran and Venetis, 2009), Twitter data, movie reviews and blogs were used to predict box office revenues for movies. In (Tumasjan et al., 2010), Twitter sentiments was also used to predict election results.
Levels of Analysis
There are basically three levels on which the sentiments analysis is performed.
1. Document Level- It was described ( Pang, Lee and Vaithyanathan, 2002; Turney, 2002) as to classify whether a whole document gives a positive or negative opinion. This is Document level Analysis. For example- A document having review of a product will enable system to tell whether the overview of the product’s review is positive or negative. However, this method only allows the analysis of a single product. It cannot analyze documents containing comparison of different products.
2. Sentence Level- The analysis is done on the sentence level. The system decides the positive, negative or neutral feedback of every sentence given as the input. It analyzes the sentence based on various rules formulated by many researchers to understand the natural language used by people.
3. Entity and Aspect Level- Entity is the target product about which the opinion is given. It is also referred to as Opinion Target. It is really necessary to know the opinion target along with the sentiment attached to it. As the sentiment alone is not enough for analysis. Aspect of an entity is the features of the given product. So the main goal of this level is to extract the sentiments or opinions on entities or their aspects. For example- “Lenovo has a really fast processor but has low battery life” Here, the aspect is the Lenovo’s Processor and battery. The opinion is positive for the processor but negative for the battery.
There are also two types of Opinions (Jindal and Liu, 2006)-
1. Regular Opinions- These opinions target a single aspect of the entity.
2. Comparative Opinions- These opinions compares the aspect/entity with some other aspect/entity.
Sentiment Lexicon
Sentiment Lexicons or Opinion Lexicons are the list of sentiment words which helps to indicate the positive or negative review. These are certainly the most important part of the sentence to deal with. Positive Lexicons are ‘good’, ‘amazing’ and negative lexicons are ‘bad’, ‘pathetic’, etc. There are various algorithms designed to extract sentiment lexicons and indicate a suitable sentiment. But there are various issues where these algorithms fail. Some issues are-
• Sarcastic sentences are difficult to deal with. Since only extracting sentiment lexicon and indicating its sentiment would make it completely wrong. In order to solve this, we need the system to understand the whole context of the sentence.
• Some sentences have sentiment words but do not express any opinion. Such sentences could be interrogative or conditional sentences. For example- “Can you suggest me a good Samsung phone?” Even though it has sentiment word like ‘good’ but it does not indicate an opinion.
• Some sentiment lexicons have different interpretation in different context. For example- “The movie was too slow and too long” and “The noodle is too long”. Here ‘long’ though used for the indicating the same meaning has different sentiment attached.
• There are sentences which have no sentiment lexicons and yet represent an opinion. For example- “This bike uses a lot of fuel”. The sentence here indicates a negative opinion about the bike but does not have any sentiment words to indicate it.
Online Spam Detection
Internet is full of reviews about everything today. It is now the primary source for getting the opinions about any product, service or company. It influences the public’s opinion and manipulate them. If people find that a particular restaurant has a high rating, then they will definitely order food from it when compared to a low-rating restaurant. So basically they play an important role for any organization and thus they need to maintain their reputation on the internet as well. This brings up the issue of Spam. Companies sometimes uses unethical methods to get a better rating such as fake reviews. These cases have been reported in the past.
Individuals maintain their privacy while giving their opinions on the internet which makes them feel secure. But some people use this freedom for their own malicious benefits. They put fake reviews about other products or services and lower its rating. Such activities are “Online Spamming” and such people are called “Online Spammers” (Jindal and Liu, 2008). It becomes a major issue of sorting out the fake reviews among the large amount of reviews given on the internet. As these reviews appear to be as legit as any other review present. This problem needs to be dealt with otherwise the platform of internet to get one’s opinion will be considered unreliable.